An application of particle filter for FDI oriented change detection and bounded parameter estimation

In their original formulations, state estimation schemes such as Kalman Filter, do not allow the incorporation of prior information on their physical bounds. This results in a certain probability of generating estimates that are physically impossible. Also, the Gaussian assumption in conventional schemes produces a trade-off between estimation error and estimation speed. This paper presents a solution based on a particle filter for which a bounded a priori parameter distribution is chosen. It is shown that a Beta distribution with hard bounds and adaptive estimation variance can overcome both drawbacks, thus achieving a lower fault detection time delay without increasing the estimation error, compared with the Extended Kalman Filter.